Monitoring the laser welding process is crucial for understanding the welding status and improving the weld quality. This paper focuses on reconstructing the keyhole and predicting the laser-beam absorptance by observing only the upper side of the specimen using one optical camera during the full-penetration multi-mode fiber laser keyhole welding of aluminum (Al5052-H32). Both the top and bottom apertures of the keyhole could be observed coaxially from above, allowing for this method to be used. Two processes were utilized to predict absorptance: object detection (YOLOR-CSP) and CNN regression (ResNet). To enhance interpretability, the average absorptance computed from the instantaneous absorptance was used. The predicted absorptance was compared with results obtained by employing a two-camera based monitoring method [1, 2], which is considered more accurate as top and bottom surfaces were observed separately by two different cameras. It was found that a comparable level of monitoring accuracy can be achieved even when using only one camera instead of two. It was also confirmed that changes in welding mode and defects can be detected well using only one camera.

[1] Analysis of laser-beam absorptance and keyhole behavior during laser keyhole welding of aluminum alloy using a deep-learning-based monitoring system, H. Kim, K. Nam, Y. Kim, H. Ki, Journal of Manufacturing Processes 80, 75-86 (2022).

[2] Deep-learning-based real-time monitoring of full-penetration laser keyhole welding using the synchronized coaxial observation method, H. Kim, K. Nam, S. Oh and H. Ki, Journal of Manufacturing Processes 68, 1018-1030 (2021).